Will A.I. Kill Us All? Our Guest Says It Might. | Interview

Quick Overview

Eleazar Yudkowsky argues that the development of Artificial General Intelligence (AGI) or superhuman AI poses an existential threat to humanity, asserting that current alignment technology is insufficient to control such systems, and thus, building them will inevitably lead to human extinction, proposing a global moratorium on advanced AI development as the only viable solution.

Key Points: Eleazar Yudkowsky, a leading voice in the AI risk movement and founder of MIRI, asserts that building superhuman AI will inevitably lead to human extinction because we lack the technology to ensure such systems are benevolent. Yudkowsky warns that powerful AI, even if not malicious, could destroy humanity as a side effect of pursuing its own goals, citing the paperclip maximizer thought experiment as an analogy for an AI indifferent to human survival while optimizing for a narrow objective. He criticizes the notion that current AI progress, like LLMs, offers hope, stating these are different technological tracks and that the failure of current alignment technology, evidenced by AI-assisted suicide incidents, highlights the impossibility of controlling superintelligence. Yudkowsky proposes a global moratorium on advanced AI development, drawing parallels to nuclear non-proliferation, advocating for an international supervisory regime over AI chips and data centers to halt capability escalation. He dismisses arguments that AI doomerism is mere hype benefiting AI companies, deeming it historically false and analogous to denying the dangers of leaded gasoline or nuclear weapons, stating that the existential threat is a scientific reality independent of market forces. Yudkowsky believes that the only way to avert disaster is through international cooperation and treaties, comparing the required mobilization to that of World War II, and emphasizes that individual acts of violence against AI researchers will not prevent global extinction.

Context: The video features an interview with Eleazar Yudkowsky, a prominent figure in the AI risk movement, known for founding the Machine Intelligence Research Institute (MIRI) and for his early warnings about existential threats from AI. Yudkowsky, who was instrumental in the early AI scene and influenced figures like Sam Altman and DeepMind's founders, now advocates for halting the development of superhuman AI. He has a new book, "If Anyone Builds It, Everyone Dies," co-authored with Nate Sores, which popularizes his long-standing arguments about the dangers of AGI. The interview also touches upon the current AI landscape, including tools like ChatGPT, and the broader debate around AI timelines and safety.